Firing a run
There are two ways a chain runs:
Run chain on the chain's page fires it immediately.
A chain with Run on a schedule turned on fires itself on its cadence.
Each fire runs the whole chain once per attached client, and you'll get a notification when it settles — completed or failed.
Following along
Chains report progress honestly — you see the real step counter, not a spinning guess:
The Runs rail on the chain's page lists every run with its status, trigger (manual or scheduled), and a live per-step view: queued, running with elapsed time, or done with the finished duration.
On the Workflow board, each fire lands as one bundled card per client, badged Chain (or Chain · scheduled). The card shows a dot per step and a line like "Step 2 of 3" with the agent currently working. It sits in Running while in flight and moves to Review when it settles.
Long steps are normal — some agents work for many minutes. If a run truly stops moving for a long stretch, the rail says so plainly and suggests running the chain again.
Note: Brief upstream hiccups are retried automatically before a step is declared failed — a single flaky moment usually won't sink the run.
Reviewing the package
Open a settled run in the rail (or the board card's Steps panel) and expand any step to see:
Its full output.
Its files as download chips.
A link to open the step's chat — the complete working transcript.
The finished deliverable on the board card is the final step's output, and the package collects the files from every step along the way — so the plan, the build sheet, and the review all arrive as one set.
When a run fails
Failed runs aren't hidden — the card still lands in Review, marked failed at the step that broke. The failed step opens already expanded with its error and chat link, so you can see exactly what happened before deciding what to do next.
Revising at step level
You don't have to rerun everything to fix one link. On any settled run, choose Revise with notes…:
Pick the scope — Whole chain (fresh run) or From step N, restarting at the step you name.
Write what should change; your notes ride along as reviewer context.
Submit. A whole-chain revise starts a brand-new run (the original is kept for comparison); a step revise resumes the same run at that step, keeping every earlier step's output.
Tip: Step-level revise is the fast path when the plan was right but the build missed the mark — rerun from the build step with a pointed note, and the chain carries the good work forward.
Only the chain's owner can revise; anyone can run. See "What are agent chains" for how chains are built and shared.